[ofa-general] Re: [PATCH 0/9 Rev3] Implement batching skb API and support in IPoIB

Krishna Kumar2 krkumar2 at in.ibm.com
Mon Sep 3 02:21:21 PDT 2007


Hi Dave,

David Miller <davem at davemloft.net> wrote on 08/29/2007 10:21:50 AM:

> From: Krishna Kumar2 <krkumar2 at in.ibm.com>
> Date: Wed, 29 Aug 2007 08:53:30 +0530
>
> > I am scp'ng from 192.168.1.1 to 192.168.1.2 and captured at the send
> > side.
>
> Bad choice of test, this is cpu limited since the scp
> has to encrypt and MAC hash all the data it sends.
>
> Use something like straight ftp or "bw_tcp" from lmbench.

I used bw_tcp from lmbench-3. I transfered 500MB and captured
the tcpdump, and analysis at various points gave pipeline sizes:

26064, 27792, 22888, 23168, 23448, 20272, 23168, 4344, 10136,
164792, 35920, 26344, 24336, 24336, 23168, 25784, 23168,

There was one huge 164K, otherwise most were in smaller ranges
like 20-30K. I ran the following test script:

SERVER=192.168.1.2
BYTES=100m
BUFFERSIZES="4096 16384 65536 131072 262144"
PROCS="1 8"
ITERATIONS=5

for m in $BUFFERSIZES
do
      for procs in $PROCS
      do
            echo TEST: Size:$m Procs:$procs
            bw_tcp -N $ITERATIONS -m $m -M $BYTES -P $procs $SERVER
      done
done

Result is:
Test without batching:
#       Size       Procs     BW (MB/s)
1       4096       1         117.39
2       16384      1         117.49
3       65536      1         117.55
4       131072     1         117.55
5       262144     1         117.58

6       4096       8         117.18
7       16384      8         117.47
8       65536      8         117.54
9       131072     8         117.59
10      262144     8         117.55

Test with batching:
#       Size       Procs     BW (MB/s)
1       4096       1         117.39
2       16384      1         117.48
3       65536      1         117.55
4       131072     1         117.58
5       262144     1         117.58

6       4096       8         117.19
7       16384      8         117.46
8       65536      8         117.53
9       131072     8         117.55
10      262144     8         117.60

So it doesn't seem to harm e1000.

Can someone give a link to the E1000E driver? I couldn't find it
after downloading Jeff's netdev-2.6 tree.

Thanks,

- KK




More information about the general mailing list